[PATCH] x86_64: Add log1p with FMA

[Sunil Pandey]

On Mon, Aug 21, 2023 at 5:32 AM Paul Zimmermann via Libc-alpha <
libc-alpha@sourceware.org> wrote:

>        Hi H.J.
>
> > On Skylake, it changes log1p bench performance by:
> >
> >         Before       After     Improvement
> > max     63.349       58.347       8%
> > min     4.448        5.651        -30%
> > mean    12.0674      10.336       14%
>
> the largest known error decreases from 0.903 ulp:
>
> log1p 0 -1 -0x1.2c10396268852p-2 [1] [0.903] 0.902709 0.9027087876691803
>
> to 0.899 ulp:
>
> log1p 0 -1 -0x1.2bf183e0344b2p-2 [1] [0.899] 0.898172 0.8981716806917975
>
> LGTM.
>
> Paul
>
> PS: the largest error for asinh is unchanged:
> asinh 0 -1 -0x1.02657ff36d5f3p-2 [2] [1.92] 1.91356 1.913551236624182
> and for atanh it seems to improve.
>

 I would like to backport this patch to release branches.

Any comments or objections?

--Sunil
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://sourceware.org/pipermail/libc-alpha/attachments/20241212/173adcec/attachment.htm>